基于局部多项式回归去势的非线性趋势序列的单位根检验

Unit Root Test of Series with Nonlinear Trend Based on Local Polynomial Regression

本文首先利用局部多项式回归来去除确定性趋势,然后利用残差差分序列长短时方差比进行单位根检验。该方法不用考虑趋势的具体形式及设定问题,因而也就回避了对趋势设定的检验问题。其次,本文研究了检验统计量的极限分布,仿真研究了窗宽的选择问题,以及残差相关与不相关时的检验水平与检验功效。最后,检验结果表明该方法是有效的。

This paper proposes an approach of unit root test of residual series after detrending deterministic trend through local polynomial regression. The deterministic trend may be in any forms including linear or nonlinear one, as well as in form of non-trend. It is not necessary to identify the sort and the specification of the trend in advance, therefore it is able to circumvent the test of the trend specification. This paper develops the limiting distribution of unit root test based on the ratio of long and short run variance of the difference of residual series, and researches how to select the bandwidth by Monte Carlo and how to simulates the test size and power with correlated and uncorrelated residual series. The result demonstrates that the approach is effective.